On Wed, 26 Apr 2006 22:05:40 +0200 (CEST) Guennadi Liakhovetski wrote: > Hi > > On Tue, 25 Apr 2006, Randy.Dunlap wrote: > > > On Tue, 25 Apr 2006 22:41:28 +0200 (CEST) Guennadi Liakhovetski wrote: > > > > > On Tue, 25 Apr 2006, Randy.Dunlap wrote: > > > > > > > I've spent a lot of time using the sg-PIO patch with the aha152x > > > > driver. The sg-PIO part of it works for me, but the aha152x driver > > > > sometimes does not complete a full DMA-out transfer (data-out phase). > > > > The driver calls this the need to "resend" some data (in datao_end()). > > > > > > Sorry, didn't quite understand it - you mean the "in" direction modified > > > to use sg-pio works, whereas the "out" doesn't? How exactly it doesn't > > > work? > > > > It's not a problem with the (your) sg patch. It's just that this > > SCSI chip sometimes (apparently) stops a DMA-out transfer before > > it should (e.g., half way thru an sg list). The driver detects this > > and decides that it needs to "resend" the remaining part of the sg > > list, but first it must back up the sg list ptr (possibly), the data > > transfer ptr (before this patch), this_residual, and possibly > > buffers_residual. I've made many attempts at doing that, but > > they are not succeeding. > > What kind of device is it and what are you doing to reproduce the error? > Do you know if the stock kernel / aha152x driver work if you boot without > highmem? I'm using a PCMCIA (not CardBus) Adapter AP-1460 card on a new Dell notebook, with an iomega Jaz drive attached to it. I haven't tested without highmem. I know that the in-kernel driver with a one-line patch works for me. The patch forces all buffers to lowmem, by setting + .unchecked_isa_dma = 1, in the scsi_host_template. I can test with a non-highmem kernel if you want me to, but I don't see much need to do that. To reproduce the error: a kernel build will do it. However, the easiest way (hardest on the driver) is: mkfs.ext3 /dev/sdX4 > Transferring "residual" bytes left after an incomplete DMA is exactly the > same situation as in dc395x and tmscsim... > > > > Could you send me / upload your current patch using k(un)map_atomic_pio? > > > > Sure, it's here: > > http://www.xenotime.net/linux/patches/aha152x-sg-pio.patch > > Thanks, looking at it. I have an old ISA SCSI card, that runs under > aha152x... --- ~Randy - : send the line "unsubscribe linux-scsi" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
